Counterfactual Estimation – Relationship Duration Models
Emily Pollock
11/4/2019
model estimation using ergm.ego
Marriage/Cohab Network
note that these models don’t include a condition for degree in other network
Fit w/ Nodecov(age)
Pop = 5,000
fit.marcoh.nodecov <- ergm.ego(egodat_marcoh ~ edges +
nodecov("age") +
absdiff("sqrtage") +
offset(nodematch("male", diff = FALSE)) +
offset("concurrent"),
offset.coef = c(-Inf, -Inf),
control = control.ergm.ego(ppopsize = 5000))## Warning in ergm.ego(egodat_marcoh ~ edges + nodecov("age") +
## absdiff("sqrtage") + : Using a smaller pseudopopulation size than sample
## size usually does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 996, different from requested 5000. Estimation should not be meaningfully affected.## Warning: `set_attrs()` is deprecated as of rlang 0.3.0
## This warning is displayed once per session.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 20:## Optimizing with step length 0.150042235315009.## The log-likelihood improved by 3.986.## Iteration 2 of at most 20:## Optimizing with step length 0.184768394150753.## The log-likelihood improved by 3.918.## Iteration 3 of at most 20:## Optimizing with step length 0.241719684674603.## The log-likelihood improved by 3.803.## Iteration 4 of at most 20:## Optimizing with step length 0.318860018061737.## The log-likelihood improved by 4.406.## Iteration 5 of at most 20:## Optimizing with step length 0.473848799904041.## The log-likelihood improved by 3.96.## Iteration 6 of at most 20:## Optimizing with step length 0.803573190792369.## The log-likelihood improved by 2.934.## Iteration 7 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.1162.## Step length converged once. Increasing MCMC sample size.## Iteration 8 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.02746.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.saveRDS(fit.marcoh.nodecov, "~/Documents/Dissertation/R/Duration/Setup/fits/marcoh-nodecov.RDS")
fit.marcoh.nodecov.5000 <- readRDS("~/Documents/Dissertation/R/Duration/Setup/fits/marcoh-nodecov.RDS")
summary(fit.marcoh.nodecov.5000)##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_marcoh ~ edges + nodecov("age") + absdiff("sqrtage") +
## offset(nodematch("male", diff = FALSE)) + offset("concurrent")
##
## Iterations: 8 out of 20
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -6.903747 0.000000 0 -Inf <1e-04 ***
## edges 2.637319 0.094971 2 27.770 <1e-04 ***
## nodecov.age 0.001103 0.001487 1 0.742 0.458
## absdiff.sqrtage -3.051756 0.039284 3 -77.685 <1e-04 ***
## offset(nodematch.male) -Inf 0.000000 0 -Inf <1e-04 ***
## offset(concurrent) -Inf 0.000000 0 -Inf <1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male) offset(concurrent)mcmc.diagnostics(fit.marcoh.nodecov.5000)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges -0.3067 9.101 0.14220 0.3185
## nodecov.age -24.8586 600.348 9.38043 21.0944
## absdiff.sqrtage -1.0553 4.937 0.07714 0.1430
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -18.03 -7.025 -0.02524 5.975 16.975
## nodecov.age -1171.25 -449.002 -20.75164 387.498 1140.623
## absdiff.sqrtage -10.85 -4.394 -0.91784 2.420 8.337
##
##
## Sample statistics cross-correlations:
## edges nodecov.age absdiff.sqrtage
## edges 1.0000000 0.9663941 0.4906161
## nodecov.age 0.9663941 1.0000000 0.4581680
## absdiff.sqrtage 0.4906161 0.4581680 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodecov.age absdiff.sqrtage
## Lag 0 1.0000000 1.0000000 1.00000000
## Lag 1024 0.5856760 0.5946789 0.33310408
## Lag 2048 0.3996049 0.4076548 0.19314507
## Lag 3072 0.2959090 0.2984055 0.15922277
## Lag 4096 0.2163162 0.2174221 0.11174744
## Lag 5120 0.1529273 0.1494340 0.08566264
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodecov.age absdiff.sqrtage
## 0.5401 0.1175 0.3588
##
## Individual P-values (lower = worse):
## edges nodecov.age absdiff.sqrtage
## 0.5891228 0.9064620 0.7197547
## Joint P-value (lower = worse): 0.4514442 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).plot(gof(fit.marcoh.nodecov.5000, GOF = "model"))
plot(gof(fit.marcoh.nodecov.5000, GOF = "degree"))
Pop = 10,000
fit.marcoh.nodecov.10 <- ergm.ego(egodat_marcoh ~ edges +
nodecov("age") +
absdiff("sqrtage") +
offset(nodematch("male", diff = FALSE)) +
offset("concurrent"),
offset.coef = c(-Inf, -Inf),
control = control.ergm.ego(ppopsize = 10000))## Warning in ergm.ego(egodat_marcoh ~ edges + nodecov("age") +
## absdiff("sqrtage") + : Using a smaller pseudopopulation size than sample
## size usually does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 5006, different from requested 10000. Estimation should not be meaningfully affected.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 20:## Optimizing with step length 0.0591582553253478.## The log-likelihood improved by 3.474.## Iteration 2 of at most 20:## Optimizing with step length 0.0648223360165142.## The log-likelihood improved by 3.797.## Iteration 3 of at most 20:## Optimizing with step length 0.0769241182426149.## The log-likelihood improved by 3.914.## Iteration 4 of at most 20:## Optimizing with step length 0.0889819306858537.## The log-likelihood improved by 3.824.## Iteration 5 of at most 20:## Optimizing with step length 0.0907723196496665.## The log-likelihood improved by 3.552.## Iteration 6 of at most 20:## Optimizing with step length 0.124817502605916.## The log-likelihood improved by 4.689.## Iteration 7 of at most 20:## Optimizing with step length 0.12169086622455.## The log-likelihood improved by 3.093.## Iteration 8 of at most 20:## Optimizing with step length 0.143899982757126.## The log-likelihood improved by 3.85.## Iteration 9 of at most 20:## Optimizing with step length 0.143084250766046.## The log-likelihood improved by 3.208.## Iteration 10 of at most 20:## Optimizing with step length 0.158818583800834.## The log-likelihood improved by 2.763.## Iteration 11 of at most 20:## Optimizing with step length 0.205505648187867.## The log-likelihood improved by 2.865.## Iteration 12 of at most 20:## Optimizing with step length 0.310967888679275.## The log-likelihood improved by 3.357.## Iteration 13 of at most 20:## Optimizing with step length 0.408100129010225.## The log-likelihood improved by 3.375.## Iteration 14 of at most 20:## Optimizing with step length 0.612143534458838.## The log-likelihood improved by 4.049.## Iteration 15 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 1.631.## Step length converged once. Increasing MCMC sample size.## Iteration 16 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.05173.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.saveRDS(fit.marcoh.nodecov.10, "~/Documents/Dissertation/R/Duration/Setup/fits/marcoh-nodecov-10.RDS")
fit.marcoh.nodecov.10 <- readRDS("~/Documents/Dissertation/R/Duration/Setup/fits/marcoh-nodecov-10.RDS")
summary(fit.marcoh.nodecov.10)##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_marcoh ~ edges + nodecov("age") + absdiff("sqrtage") +
## offset(nodematch("male", diff = FALSE)) + offset("concurrent")
##
## Iterations: 16 out of 20
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -8.518392 0.000000 0 -Inf <1e-04 ***
## edges 0.867611 0.107857 1 8.044 <1e-04 ***
## nodecov.age 0.030163 0.001748 1 17.256 <1e-04 ***
## absdiff.sqrtage -3.083225 0.041778 1 -73.800 <1e-04 ***
## offset(nodematch.male) -Inf 0.000000 0 -Inf <1e-04 ***
## offset(concurrent) -Inf 0.000000 0 -Inf <1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male) offset(concurrent)mcmc.diagnostics(fit.marcoh.nodecov.10)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges -4.460 19.20 0.3000 1.414
## nodecov.age -347.269 1241.79 19.4030 101.744
## absdiff.sqrtage -1.428 10.47 0.1635 0.532
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -40.83 -17.201 -5.201 7.799 33.80
## nodecov.age -2695.86 -1163.858 -366.858 458.392 2197.39
## absdiff.sqrtage -23.22 -8.319 -1.146 5.765 17.86
##
##
## Sample statistics cross-correlations:
## edges nodecov.age absdiff.sqrtage
## edges 1.0000000 0.9655889 0.4434295
## nodecov.age 0.9655889 1.0000000 0.3987311
## absdiff.sqrtage 0.4434295 0.3987311 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodecov.age absdiff.sqrtage
## Lag 0 1.0000000 1.0000000 1.0000000
## Lag 1024 0.8720926 0.8901068 0.7558804
## Lag 2048 0.7696055 0.8003449 0.5870290
## Lag 3072 0.6911096 0.7322043 0.4736310
## Lag 4096 0.6236552 0.6726041 0.3943862
## Lag 5120 0.5744464 0.6292782 0.3427052
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodecov.age absdiff.sqrtage
## -0.2886 -0.2572 0.2044
##
## Individual P-values (lower = worse):
## edges nodecov.age absdiff.sqrtage
## 0.7728602 0.7969877 0.8380190
## Joint P-value (lower = worse): 0.985404 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).plot(gof(fit.marcoh.nodecov.10, GOF = "model"))
plot(gof(fit.marcoh.nodecov.10, GOF = "degree"))
Pop = 20,000
fit.marcoh.nodecov.20 <- ergm.ego(egodat_marcoh ~ edges +
nodecov("age") +
absdiff("sqrtage") +
offset(nodematch("male", diff = FALSE)) +
offset("concurrent"),
offset.coef = c(-Inf, -Inf),
control = control.ergm.ego(ppopsize = 20000, ergm.control = control.ergm(MCMLE.maxit = 100)))## Warning in ergm.ego(egodat_marcoh ~ edges + nodecov("age") +
## absdiff("sqrtage") + : Using a smaller pseudopopulation size than sample
## size usually does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 15250, different from requested 20000. Estimation should not be meaningfully affected.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 100:## Optimizing with step length 0.0342073531590534.## The log-likelihood improved by 3.651.## Iteration 2 of at most 100:## Optimizing with step length 0.0368780104748084.## The log-likelihood improved by 4.135.## Iteration 3 of at most 100:## Optimizing with step length 0.0331072743074978.## The log-likelihood improved by 2.085.## Iteration 4 of at most 100:## Optimizing with step length 0.0383696071667242.## The log-likelihood improved by 3.347.## Iteration 5 of at most 100:## Optimizing with step length 0.033236621402204.## The log-likelihood improved by 2.554.## Iteration 6 of at most 100:## Optimizing with step length 0.0351892707955161.## The log-likelihood improved by 2.568.## Iteration 7 of at most 100:## Optimizing with step length 0.0514152205687653.## The log-likelihood improved by 3.943.## Iteration 8 of at most 100:## Optimizing with step length 0.043573178023531.## The log-likelihood improved by 3.118.## Iteration 9 of at most 100:## Optimizing with step length 0.0549693383362363.## The log-likelihood improved by 2.957.## Iteration 10 of at most 100:## Optimizing with step length 0.0431393792190266.## The log-likelihood improved by 2.91.## Iteration 11 of at most 100:## Optimizing with step length 0.0459161773889357.## The log-likelihood improved by 2.291.## Iteration 12 of at most 100:## Optimizing with step length 0.0520037642683399.## The log-likelihood improved by 3.209.## Iteration 13 of at most 100:## Optimizing with step length 0.0619969935861979.## The log-likelihood improved by 2.462.## Iteration 14 of at most 100:## Optimizing with step length 0.0678492998688997.## The log-likelihood improved by 2.736.## Iteration 15 of at most 100:## Optimizing with step length 0.0567557691487198.## The log-likelihood improved by 2.476.## Iteration 16 of at most 100:## Optimizing with step length 0.0881004127452315.## The log-likelihood improved by 4.344.## Iteration 17 of at most 100:## Optimizing with step length 0.0959876744419226.## The log-likelihood improved by 3.6.## Iteration 18 of at most 100:## Optimizing with step length 0.0905414610044193.## The log-likelihood improved by 2.931.## Iteration 19 of at most 100:## Optimizing with step length 0.135727782876339.## The log-likelihood improved by 4.055.## Iteration 20 of at most 100:## Optimizing with step length 0.117085639980348.## The log-likelihood improved by 3.22.## Iteration 21 of at most 100:## Optimizing with step length 0.144280693365455.## The log-likelihood improved by 3.25.## Iteration 22 of at most 100:## Optimizing with step length 0.176862066242742.## The log-likelihood improved by 2.712.## Iteration 23 of at most 100:## Optimizing with step length 0.189311755049649.## The log-likelihood improved by 2.948.## Iteration 24 of at most 100:## Optimizing with step length 0.262724580666936.## The log-likelihood improved by 2.885.## Iteration 25 of at most 100:## Optimizing with step length 0.253378759449056.## The log-likelihood improved by 2.439.## Iteration 26 of at most 100:## Optimizing with step length 0.372049530334089.## The log-likelihood improved by 2.17.## Iteration 27 of at most 100:## Optimizing with step length 0.691218460665044.## The log-likelihood improved by 2.305.## Iteration 28 of at most 100:## Optimizing with step length 1.## The log-likelihood improved by 0.6534.## Step length converged once. Increasing MCMC sample size.## Iteration 29 of at most 100:## Optimizing with step length 1.## The log-likelihood improved by 0.5366.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.saveRDS(fit.marcoh.nodecov.20, "~/Documents/Dissertation/R/Duration/Setup/fits/marcoh-nodecov-20.RDS")
fit.marcoh.nodecov.20 <- readRDS("~/Documents/Dissertation/R/Duration/Setup/fits/marcoh-nodecov-20.RDS")
summary(fit.marcoh.nodecov.20)##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_marcoh ~ edges + nodecov("age") + absdiff("sqrtage") +
## offset(nodematch("male", diff = FALSE)) + offset("concurrent")
##
## Iterations: 29 out of 100
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -9.632335 0.000000 0 -Inf <1e-04 ***
## edges 0.052876 0.113834 1 0.464 0.642
## nodecov.age 0.044424 0.001927 1 23.049 <1e-04 ***
## absdiff.sqrtage -3.073709 0.062027 1 -49.554 <1e-04 ***
## offset(nodematch.male) -Inf 0.000000 0 -Inf <1e-04 ***
## offset(concurrent) -Inf 0.000000 0 -Inf <1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male) offset(concurrent)mcmc.diagnostics(fit.marcoh.nodecov.20)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges -6.92 33.81 0.5282 3.556
## nodecov.age -238.12 2155.49 33.6795 252.337
## absdiff.sqrtage 10.62 16.70 0.2609 1.107
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -72.73 -30.734 -5.734 17.27 57.27
## nodecov.age -4521.99 -1767.735 -157.235 1355.26 3758.01
## absdiff.sqrtage -20.91 -1.059 10.409 22.47 42.67
##
##
## Sample statistics cross-correlations:
## edges nodecov.age absdiff.sqrtage
## edges 1.0000000 0.9686980 0.6071024
## nodecov.age 0.9686980 1.0000000 0.5907839
## absdiff.sqrtage 0.6071024 0.5907839 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodecov.age absdiff.sqrtage
## Lag 0 1.0000000 1.0000000 1.0000000
## Lag 1024 0.9538279 0.9626760 0.8946665
## Lag 2048 0.9128547 0.9291189 0.8034705
## Lag 3072 0.8747716 0.8974731 0.7247423
## Lag 4096 0.8377046 0.8661729 0.6536490
## Lag 5120 0.8048808 0.8381512 0.5893482
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodecov.age absdiff.sqrtage
## -1.019 -0.853 -2.270
##
## Individual P-values (lower = worse):
## edges nodecov.age absdiff.sqrtage
## 0.30824083 0.39366713 0.02323372
## Joint P-value (lower = worse): 0.2219269 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).plot(gof(fit.marcoh.nodecov.20, GOF = "model"))
plot(gof(fit.marcoh.nodecov.20, GOF = "degree"))
Fit w/ Nodefactor(agecat)
Pop = 5,000
## Warning in ergm.ego(egodat_marcoh ~ edges + nodefactor("agecat", levels =
## c(1:3)) + : Using a smaller pseudopopulation size than sample size usually
## does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 996, different from requested 5000. Estimation should not be meaningfully affected.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 20:## Optimizing with step length 0.138664135039476.## The log-likelihood improved by 4.545.## Iteration 2 of at most 20:## Optimizing with step length 0.153716395299718.## The log-likelihood improved by 3.479.## Iteration 3 of at most 20:## Optimizing with step length 0.196039028519132.## The log-likelihood improved by 3.248.## Iteration 4 of at most 20:## Optimizing with step length 0.292264655747309.## The log-likelihood improved by 4.671.## Iteration 5 of at most 20:## Optimizing with step length 0.386332612059242.## The log-likelihood improved by 3.993.## Iteration 6 of at most 20:## Optimizing with step length 0.566458089107911.## The log-likelihood improved by 3.2.## Iteration 7 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 2.088.## Step length converged once. Increasing MCMC sample size.## Iteration 8 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.01264.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_marcoh ~ edges + nodefactor("agecat", levels = c(1:3)) +
## absdiff("sqrtage") + offset(nodematch("male", diff = FALSE)) +
## offset("concurrent")
##
## Iterations: 8 out of 20
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -6.90375 0.00000 0 -Inf <1e-04 ***
## edges 2.81591 0.04323 1 65.14 <1e-04 ***
## nodefactor.agecat.1 -1.39817 0.07052 1 -19.82 <1e-04 ***
## nodefactor.agecat.2 -0.44070 0.04170 1 -10.57 <1e-04 ***
## nodefactor.agecat.3 0.77052 0.07314 1 10.54 <1e-04 ***
## absdiff.sqrtage -3.01466 0.04929 3 -61.16 <1e-04 ***
## offset(nodematch.male) -Inf 0.00000 0 -Inf <1e-04 ***
## offset(concurrent) -Inf 0.00000 0 -Inf <1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male) offset(concurrent)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges 0.93398 7.996 0.12494 0.24197
## nodefactor.agecat.1 0.03991 2.775 0.04336 0.05455
## nodefactor.agecat.2 0.50836 6.569 0.10264 0.24475
## nodefactor.agecat.3 0.28465 5.375 0.08398 0.24698
## absdiff.sqrtage -0.09904 4.569 0.07140 0.16112
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -15.025 -4.025 0.97476 5.975 16.975
## nodefactor.agecat.1 -5.710 -1.710 0.29015 2.290 4.290
## nodefactor.agecat.2 -12.397 -4.397 0.60259 5.603 13.603
## nodefactor.agecat.3 -10.225 -3.225 -0.22462 3.775 10.775
## absdiff.sqrtage -9.241 -3.126 -0.08307 3.077 8.723
##
##
## Sample statistics cross-correlations:
## edges nodefactor.agecat.1 nodefactor.agecat.2
## edges 1.0000000 0.22366094 0.41540263
## nodefactor.agecat.1 0.2236609 1.00000000 0.07525687
## nodefactor.agecat.2 0.4154026 0.07525687 1.00000000
## nodefactor.agecat.3 0.3186299 -0.01691419 0.02322470
## absdiff.sqrtage 0.5322287 0.16647922 0.24091123
## nodefactor.agecat.3 absdiff.sqrtage
## edges 0.31862994 0.5322287
## nodefactor.agecat.1 -0.01691419 0.1664792
## nodefactor.agecat.2 0.02322470 0.2409112
## nodefactor.agecat.3 1.00000000 0.2035769
## absdiff.sqrtage 0.20357690 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodefactor.agecat.1 nodefactor.agecat.2
## Lag 0 1.00000000 1.00000000 1.0000000
## Lag 1024 0.53484504 0.22550351 0.5992209
## Lag 2048 0.33158932 0.05973604 0.4135625
## Lag 3072 0.20771423 0.02120235 0.2944970
## Lag 4096 0.13684739 0.02477597 0.2171836
## Lag 5120 0.09342227 0.01826479 0.1641331
## nodefactor.agecat.3 absdiff.sqrtage
## Lag 0 1.0000000 1.0000000
## Lag 1024 0.7101414 0.3926225
## Lag 2048 0.5564489 0.2537075
## Lag 3072 0.4530715 0.1738270
## Lag 4096 0.3789662 0.1688949
## Lag 5120 0.3140504 0.1434649
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodefactor.agecat.1 nodefactor.agecat.2
## 0.3424 -0.2947 1.3294
## nodefactor.agecat.3 absdiff.sqrtage
## -1.4618 0.3372
##
## Individual P-values (lower = worse):
## edges nodefactor.agecat.1 nodefactor.agecat.2
## 0.7320541 0.7682582 0.1837083
## nodefactor.agecat.3 absdiff.sqrtage
## 0.1437867 0.7359551
## Joint P-value (lower = worse): 0.5852956 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).
Pop = 10,000
## Warning in ergm.ego(egodat_marcoh ~ edges + nodefactor("agecat", levels =
## c(1:3)) + : Using a smaller pseudopopulation size than sample size usually
## does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 5006, different from requested 10000. Estimation should not be meaningfully affected.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 20:## Optimizing with step length 0.0466144665741913.## The log-likelihood improved by 2.913.## Iteration 2 of at most 20:## Optimizing with step length 0.0603402802034657.## The log-likelihood improved by 3.625.## Iteration 3 of at most 20:## Optimizing with step length 0.0599150205609965.## The log-likelihood improved by 2.821.## Iteration 4 of at most 20:## Optimizing with step length 0.0725939423739697.## The log-likelihood improved by 3.232.## Iteration 5 of at most 20:## Optimizing with step length 0.0775541246194489.## The log-likelihood improved by 3.446.## Iteration 6 of at most 20:## Optimizing with step length 0.0882368124422177.## The log-likelihood improved by 3.415.## Iteration 7 of at most 20:## Optimizing with step length 0.108669924208481.## The log-likelihood improved by 4.728.## Iteration 8 of at most 20:## Optimizing with step length 0.115116400630009.## The log-likelihood improved by 3.681.## Iteration 9 of at most 20:## Optimizing with step length 0.123601007461258.## The log-likelihood improved by 3.097.## Iteration 10 of at most 20:## Optimizing with step length 0.180630398545434.## The log-likelihood improved by 3.949.## Iteration 11 of at most 20:## Optimizing with step length 0.145712046081762.## The log-likelihood improved by 3.01.## Iteration 12 of at most 20:## Optimizing with step length 0.196711055257145.## The log-likelihood improved by 3.209.## Iteration 13 of at most 20:## Optimizing with step length 0.231854433769991.## The log-likelihood improved by 2.804.## Iteration 14 of at most 20:## Optimizing with step length 0.343516127788765.## The log-likelihood improved by 2.948.## Iteration 15 of at most 20:## Optimizing with step length 0.517242343767213.## The log-likelihood improved by 3.523.## Iteration 16 of at most 20:## Optimizing with step length 0.876462464387922.## The log-likelihood improved by 2.688.## Iteration 17 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.6248.## Step length converged once. Increasing MCMC sample size.## Iteration 18 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.213.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_marcoh ~ edges + nodefactor("agecat", levels = c(1:3)) +
## absdiff("sqrtage") + offset(nodematch("male", diff = FALSE)) +
## offset("concurrent")
##
## Iterations: 18 out of 20
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -8.51839 0.00000 0 -Inf < 1e-04 ***
## edges 3.15648 0.05056 1 62.44 < 1e-04 ***
## nodefactor.agecat.1 -1.82310 0.06772 0 -26.92 < 1e-04 ***
## nodefactor.agecat.2 -0.62319 0.04952 1 -12.59 < 1e-04 ***
## nodefactor.agecat.3 0.15856 0.06052 1 2.62 0.00879 **
## absdiff.sqrtage -3.06049 0.04478 1 -68.34 < 1e-04 ***
## offset(nodematch.male) -Inf 0.00000 0 -Inf < 1e-04 ***
## offset(concurrent) -Inf 0.00000 0 -Inf < 1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male) offset(concurrent)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges 8.3533 18.748 0.2929 1.3037
## nodefactor.agecat.1 -1.3040 6.402 0.1000 0.2252
## nodefactor.agecat.2 0.5920 14.864 0.2323 1.0072
## nodefactor.agecat.3 1.4744 14.271 0.2230 1.3126
## absdiff.sqrtage 0.1265 10.279 0.1606 0.5061
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -28.20 -4.201 8.7988 20.799 45.80
## nodefactor.agecat.1 -14.39 -5.385 -1.3850 3.615 10.61
## nodefactor.agecat.2 -27.64 -9.640 0.3600 10.360 30.36
## nodefactor.agecat.3 -26.60 -7.597 1.4032 11.403 29.40
## absdiff.sqrtage -21.12 -6.310 0.3682 6.802 19.72
##
##
## Sample statistics cross-correlations:
## edges nodefactor.agecat.1 nodefactor.agecat.2
## edges 1.0000000 0.209554247 0.52686490
## nodefactor.agecat.1 0.2095542 1.000000000 0.08133085
## nodefactor.agecat.2 0.5268649 0.081330850 1.00000000
## nodefactor.agecat.3 0.4590529 -0.002936471 0.10252669
## absdiff.sqrtage 0.4719611 0.168048541 0.22192872
## nodefactor.agecat.3 absdiff.sqrtage
## edges 0.459052880 0.4719611
## nodefactor.agecat.1 -0.002936471 0.1680485
## nodefactor.agecat.2 0.102526688 0.2219287
## nodefactor.agecat.3 1.000000000 0.2686391
## absdiff.sqrtage 0.268639115 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodefactor.agecat.1 nodefactor.agecat.2
## Lag 0 1.0000000 1.0000000 1.0000000
## Lag 1024 0.8741026 0.6702266 0.8834002
## Lag 2048 0.7797502 0.4451344 0.7882781
## Lag 3072 0.7022815 0.2980086 0.7118037
## Lag 4096 0.6411980 0.2149504 0.6440391
## Lag 5120 0.5863105 0.1600178 0.5852274
## nodefactor.agecat.3 absdiff.sqrtage
## Lag 0 1.0000000 1.0000000
## Lag 1024 0.9124824 0.7808686
## Lag 2048 0.8411940 0.6237096
## Lag 3072 0.7818052 0.5065533
## Lag 4096 0.7308717 0.4263601
## Lag 5120 0.6911989 0.3634459
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodefactor.agecat.1 nodefactor.agecat.2
## -0.4987 -1.2307 -4.7512
## nodefactor.agecat.3 absdiff.sqrtage
## 3.1402 0.7617
##
## Individual P-values (lower = worse):
## edges nodefactor.agecat.1 nodefactor.agecat.2
## 6.180045e-01 2.184344e-01 2.021785e-06
## nodefactor.agecat.3 absdiff.sqrtage
## 1.688351e-03 4.462135e-01
## Joint P-value (lower = worse): 1.378772e-06 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).
Pop = 20,000
## Warning in ergm.ego(egodat_marcoh ~ edges + nodefactor("agecat", levels =
## c(1:3)) + : Using a smaller pseudopopulation size than sample size usually
## does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 15250, different from requested 20000. Estimation should not be meaningfully affected.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 100:## Optimizing with step length 0.027873167438293.## The log-likelihood improved by 2.907.## Iteration 2 of at most 100:## Optimizing with step length 0.0323420172441747.## The log-likelihood improved by 3.661.## Iteration 3 of at most 100:## Optimizing with step length 0.0247528021364617.## The log-likelihood improved by 2.127.## Iteration 4 of at most 100:## Optimizing with step length 0.0367763085664953.## The log-likelihood improved by 4.219.## Iteration 5 of at most 100:## Optimizing with step length 0.03470863938905.## The log-likelihood improved by 3.355.## Iteration 6 of at most 100:## Optimizing with step length 0.03372154110786.## The log-likelihood improved by 2.607.## Iteration 7 of at most 100:## Optimizing with step length 0.0320368427795113.## The log-likelihood improved by 2.498.## Iteration 8 of at most 100:## Optimizing with step length 0.0390603752148347.## The log-likelihood improved by 2.619.## Iteration 9 of at most 100:## Optimizing with step length 0.0381537707928936.## The log-likelihood improved by 2.387.## Iteration 10 of at most 100:## Optimizing with step length 0.0485663680101388.## The log-likelihood improved by 3.279.## Iteration 11 of at most 100:## Optimizing with step length 0.0399393953396098.## The log-likelihood improved by 2.83.## Iteration 12 of at most 100:## Optimizing with step length 0.0463526830957157.## The log-likelihood improved by 2.612.## Iteration 13 of at most 100:## Optimizing with step length 0.0528851819311443.## The log-likelihood improved by 3.474.## Iteration 14 of at most 100:## Optimizing with step length 0.053773582757262.## The log-likelihood improved by 2.913.## Iteration 15 of at most 100:## Optimizing with step length 0.0572467238743115.## The log-likelihood improved by 2.856.## Iteration 16 of at most 100:## Optimizing with step length 0.0518456377319033.## The log-likelihood improved by 2.396.## Iteration 17 of at most 100:## Optimizing with step length 0.0703217728096083.## The log-likelihood improved by 2.403.## Iteration 18 of at most 100:## Optimizing with step length 0.0562519351018623.## The log-likelihood improved by 2.059.## Iteration 19 of at most 100:## Optimizing with step length 0.0956289477639831.## The log-likelihood improved by 4.034.## Iteration 20 of at most 100:## Optimizing with step length 0.0831273784389472.## The log-likelihood improved by 3.193.## Iteration 21 of at most 100:## Optimizing with step length 0.0885680676533335.## The log-likelihood improved by 3.28.## Iteration 22 of at most 100:## Optimizing with step length 0.0852599927827676.## The log-likelihood improved by 2.703.## Iteration 23 of at most 100:## Optimizing with step length 0.12865699754835.## The log-likelihood improved by 4.069.## Iteration 24 of at most 100:## Optimizing with step length 0.111126883118054.## The log-likelihood improved by 1.832.## Iteration 25 of at most 100:## Optimizing with step length 0.207129661926247.## The log-likelihood improved by 1.916.## Iteration 26 of at most 100:## Optimizing with step length 0.243180894809406.## The log-likelihood improved by 2.595.## Iteration 27 of at most 100:## Optimizing with step length 0.260101310836516.## The log-likelihood improved by 1.875.## Iteration 28 of at most 100:## Optimizing with step length 0.39698555147253.## The log-likelihood improved by 2.932.## Iteration 29 of at most 100:## Optimizing with step length 0.592521890725415.## The log-likelihood improved by 3.314.## Iteration 30 of at most 100:## Optimizing with step length 0.790466834696432.## The log-likelihood improved by 2.347.## Iteration 31 of at most 100:## Optimizing with step length 1.## The log-likelihood improved by 1.506.## Step length converged once. Increasing MCMC sample size.## Iteration 32 of at most 100:## Optimizing with step length 1.## The log-likelihood improved by 0.7476.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_marcoh ~ edges + nodefactor("agecat", levels = c(1:3)) +
## absdiff("sqrtage") + offset(nodematch("male", diff = FALSE)) +
## offset("concurrent")
##
## Iterations: 32 out of 100
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -9.63233 0.00000 0 -Inf <1e-04 ***
## edges 3.36280 0.05922 1 56.789 <1e-04 ***
## nodefactor.agecat.1 -2.14444 0.07060 0 -30.374 <1e-04 ***
## nodefactor.agecat.2 -0.75494 0.05460 0 -13.827 <1e-04 ***
## nodefactor.agecat.3 0.09973 0.05397 1 1.848 0.0646 .
## absdiff.sqrtage -3.09290 0.06599 1 -46.869 <1e-04 ***
## offset(nodematch.male) -Inf 0.00000 0 -Inf <1e-04 ***
## offset(concurrent) -Inf 0.00000 0 -Inf <1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male) offset(concurrent)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges 32.0200 35.32 0.5518 4.4356
## nodefactor.agecat.1 -1.7539 10.83 0.1692 0.6116
## nodefactor.agecat.2 -0.8251 26.99 0.4217 3.6071
## nodefactor.agecat.3 24.8936 28.10 0.4391 4.0713
## absdiff.sqrtage 8.5795 16.94 0.2647 1.2822
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -38.73 8.266 32.2661 57.27 97.27
## nodefactor.agecat.1 -22.69 -8.690 -1.6899 5.31 19.31
## nodefactor.agecat.2 -53.90 -18.900 -0.9001 17.10 52.72
## nodefactor.agecat.3 -25.25 3.752 22.7515 44.75 82.75
## absdiff.sqrtage -24.39 -2.906 8.2595 19.78 42.00
##
##
## Sample statistics cross-correlations:
## edges nodefactor.agecat.1 nodefactor.agecat.2
## edges 1.0000000 0.18370249 0.5944878
## nodefactor.agecat.1 0.1837025 1.00000000 0.1584014
## nodefactor.agecat.2 0.5944878 0.15840143 1.0000000
## nodefactor.agecat.3 0.5041297 -0.05212837 0.1095053
## absdiff.sqrtage 0.6493379 0.12062195 0.3810586
## nodefactor.agecat.3 absdiff.sqrtage
## edges 0.50412968 0.6493379
## nodefactor.agecat.1 -0.05212837 0.1206219
## nodefactor.agecat.2 0.10950531 0.3810586
## nodefactor.agecat.3 1.00000000 0.3318394
## absdiff.sqrtage 0.33183944 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodefactor.agecat.1 nodefactor.agecat.2
## Lag 0 1.0000000 1.0000000 1.0000000
## Lag 1024 0.9643615 0.8577625 0.9583287
## Lag 2048 0.9290293 0.7312741 0.9207521
## Lag 3072 0.8967385 0.6168457 0.8865835
## Lag 4096 0.8655751 0.5187668 0.8536933
## Lag 5120 0.8368374 0.4373850 0.8251907
## nodefactor.agecat.3 absdiff.sqrtage
## Lag 0 1.0000000 1.0000000
## Lag 1024 0.9769993 0.9182090
## Lag 2048 0.9546679 0.8422038
## Lag 3072 0.9332535 0.7725440
## Lag 4096 0.9136152 0.7083530
## Lag 5120 0.8945763 0.6515551
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodefactor.agecat.1 nodefactor.agecat.2
## -3.7935 0.8007 -0.6378
## nodefactor.agecat.3 absdiff.sqrtage
## -1.6268 -1.0726
##
## Individual P-values (lower = worse):
## edges nodefactor.agecat.1 nodefactor.agecat.2
## 0.0001485492 0.4232882134 0.5236156550
## nodefactor.agecat.3 absdiff.sqrtage
## 0.1037793669 0.2834459697
## Joint P-value (lower = worse): 0.1938805 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).
“Other” formation model:
note these models also don’t have contrainst based on degree in marcoh network
Nodecov(age)
## Warning in ergm.ego(egodat_other ~ edges + nodecov("age") +
## absdiff("sqrtage") + : Using a smaller pseudopopulation size than sample
## size usually does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 5006, different from requested 10000. Estimation should not be meaningfully affected.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 20:## Optimizing with step length 0.204291736580097.## The log-likelihood improved by 3.608.## Iteration 2 of at most 20:## Optimizing with step length 0.284888209675062.## The log-likelihood improved by 3.383.## Iteration 3 of at most 20:## Optimizing with step length 0.436574121894018.## The log-likelihood improved by 3.681.## Iteration 4 of at most 20:## Optimizing with step length 0.779438852365321.## The log-likelihood improved by 3.19.## Iteration 5 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.009037.## Step length converged once. Increasing MCMC sample size.## Iteration 6 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.003768.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_other ~ edges + nodecov("age") + absdiff("sqrtage") +
## concurrent + offset(nodematch("male", diff = FALSE))
##
## Iterations: 6 out of 20
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -8.518392 0.000000 0 -Inf <1e-04 ***
## edges 4.111824 0.107959 1 38.09 <1e-04 ***
## nodecov.age -0.061302 0.001666 1 -36.80 <1e-04 ***
## absdiff.sqrtage -2.911515 0.065375 0 -44.53 <1e-04 ***
## concurrent -1.103946 0.091461 0 -12.07 <1e-04 ***
## offset(nodematch.male) -Inf 0.000000 0 -Inf <1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges 0.1611 17.473 0.27302 0.7185
## nodecov.age 16.2576 951.540 14.86781 32.9492
## absdiff.sqrtage -0.2234 7.601 0.11876 0.2115
## concurrent 0.3120 5.949 0.09295 0.1773
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -33.19 -12.195 -0.1946 11.805 34.81
## nodecov.age -1820.37 -629.373 18.6272 658.627 1919.25
## absdiff.sqrtage -15.96 -5.194 -0.1934 4.994 14.33
## concurrent -12.17 -4.169 0.8306 4.831 11.83
##
##
## Sample statistics cross-correlations:
## edges nodecov.age absdiff.sqrtage concurrent
## edges 1.0000000 0.9641743 0.6713954 0.5105704
## nodecov.age 0.9641743 1.0000000 0.6582037 0.4407747
## absdiff.sqrtage 0.6713954 0.6582037 1.0000000 0.3250011
## concurrent 0.5105704 0.4407747 0.3250011 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodecov.age absdiff.sqrtage concurrent
## Lag 0 1.0000000 1.0000000 1.00000000 1.0000000
## Lag 1024 0.6793662 0.5976646 0.43949156 0.4407176
## Lag 2048 0.4894343 0.3822709 0.22432399 0.2574006
## Lag 3072 0.3764409 0.2645224 0.12844998 0.1590593
## Lag 4096 0.3053709 0.2003968 0.10671811 0.1208386
## Lag 5120 0.2478397 0.1501255 0.07803881 0.1044248
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodecov.age absdiff.sqrtage concurrent
## -0.53744 0.06666 -3.15729 -0.90844
##
## Individual P-values (lower = worse):
## edges nodecov.age absdiff.sqrtage concurrent
## 0.59096090 0.94685253 0.00159243 0.36364736
## Joint P-value (lower = worse): 0.00453261 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).
Nodefactor(agecat)
## Warning in ergm.ego(egodat_other ~ edges + nodefactor("agecat", levels =
## c(1:4)) + : Using a smaller pseudopopulation size than sample size usually
## does not make sense.## Constructing pseudopopulation network.## Note: Constructed network has size 5006, different from requested 10000. Estimation should not be meaningfully affected.## Unable to match target stats. Using MCMLE estimation.## Starting maximum pseudolikelihood estimation (MPLE):## Evaluating the predictor and response matrix.## Maximizing the pseudolikelihood.## Finished MPLE.## Starting Monte Carlo maximum likelihood estimation (MCMLE):## Iteration 1 of at most 20:## Optimizing with step length 0.209575703549554.## The log-likelihood improved by 3.358.## Iteration 2 of at most 20:## Optimizing with step length 0.302829116735493.## The log-likelihood improved by 2.99.## Iteration 3 of at most 20:## Optimizing with step length 0.44983110233985.## The log-likelihood improved by 3.188.## Iteration 4 of at most 20:## Optimizing with step length 0.903618676598601.## The log-likelihood improved by 3.733.## Iteration 5 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.5363.## Step length converged once. Increasing MCMC sample size.## Iteration 6 of at most 20:## Optimizing with step length 1.## The log-likelihood improved by 0.09736.## Step length converged twice. Stopping.## Finished MCMLE.## This model was fit using MCMC. To examine model diagnostics and
## check for degeneracy, use the mcmc.diagnostics() function.##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: egodat_other ~ edges + nodefactor("agecat", levels = c(1:4)) +
## absdiff("sqrtage") + concurrent + offset(nodematch("male",
## diff = FALSE))
##
## Iterations: 6 out of 20
##
## Monte Carlo MLE Results:
## Estimate Std. Error MCMC % z value Pr(>|z|)
## offset(netsize.adj) -8.51839 0.00000 0 -Inf <1e-04 ***
## edges -0.64490 0.04943 0 -13.046 <1e-04 ***
## nodefactor.agecat.1 1.34147 0.04047 1 33.147 <1e-04 ***
## nodefactor.agecat.2 1.10352 0.03876 1 28.467 <1e-04 ***
## nodefactor.agecat.3 0.66361 0.03504 1 18.938 <1e-04 ***
## nodefactor.agecat.4 0.34566 0.04200 0 8.229 <1e-04 ***
## absdiff.sqrtage -2.88201 0.06721 0 -42.881 <1e-04 ***
## concurrent -1.14399 0.08878 0 -12.886 <1e-04 ***
## offset(nodematch.male) -Inf 0.00000 0 -Inf <1e-04 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
## The following terms are fixed by offset and are not estimated:
## offset(netsize.adj) offset(nodematch.male)## Sample statistics summary:
##
## Iterations = 16384:4209664
## Thinning interval = 1024
## Number of chains = 1
## Sample size per chain = 4096
##
## 1. Empirical mean and standard deviation for each variable,
## plus standard error of the mean:
##
## Mean SD Naive SE Time-series SE
## edges 1.2819 17.294 0.27022 0.8176
## nodefactor.agecat.1 2.6692 13.940 0.21782 1.1355
## nodefactor.agecat.2 -4.0351 17.525 0.27382 0.9862
## nodefactor.agecat.3 1.4807 14.011 0.21892 0.6572
## nodefactor.agecat.4 1.9532 11.027 0.17229 0.3764
## absdiff.sqrtage -0.3529 7.691 0.12017 0.2392
## concurrent -0.8896 6.100 0.09532 0.2299
##
## 2. Quantiles for each variable:
##
## 2.5% 25% 50% 75% 97.5%
## edges -34.19 -10.195 1.8054 12.805 33.805
## nodefactor.agecat.1 -26.91 -6.913 3.0874 13.087 27.087
## nodefactor.agecat.2 -39.09 -15.094 -4.0937 7.906 30.906
## nodefactor.agecat.3 -27.27 -7.643 1.3572 11.357 27.982
## nodefactor.agecat.4 -21.21 -5.210 2.7896 9.790 22.790
## absdiff.sqrtage -15.85 -5.295 -0.1209 4.843 14.526
## concurrent -13.17 -5.169 -1.1694 3.831 9.831
##
##
## Sample statistics cross-correlations:
## edges nodefactor.agecat.1 nodefactor.agecat.2
## edges 1.0000000 0.471269288 0.58871507
## nodefactor.agecat.1 0.4712693 1.000000000 0.09816441
## nodefactor.agecat.2 0.5887151 0.098164412 1.00000000
## nodefactor.agecat.3 0.4866158 0.043884665 0.07085743
## nodefactor.agecat.4 0.4100871 -0.001948945 0.05160447
## absdiff.sqrtage 0.6886763 0.320036234 0.38510857
## concurrent 0.4816460 0.298744217 0.43518962
## nodefactor.agecat.3 nodefactor.agecat.4
## edges 0.48661575 0.410087088
## nodefactor.agecat.1 0.04388467 -0.001948945
## nodefactor.agecat.2 0.07085743 0.051604472
## nodefactor.agecat.3 1.00000000 0.067289692
## nodefactor.agecat.4 0.06728969 1.000000000
## absdiff.sqrtage 0.36007275 0.293847899
## concurrent 0.21452611 0.105806034
## absdiff.sqrtage concurrent
## edges 0.6886763 0.4816460
## nodefactor.agecat.1 0.3200362 0.2987442
## nodefactor.agecat.2 0.3851086 0.4351896
## nodefactor.agecat.3 0.3600728 0.2145261
## nodefactor.agecat.4 0.2938479 0.1058060
## absdiff.sqrtage 1.0000000 0.3135916
## concurrent 0.3135916 1.0000000
##
## Sample statistics auto-correlation:
## Chain 1
## edges nodefactor.agecat.1 nodefactor.agecat.2
## Lag 0 1.0000000 1.0000000 1.0000000
## Lag 1024 0.6686986 0.8681123 0.8261418
## Lag 2048 0.4961538 0.7819380 0.6996014
## Lag 3072 0.4034751 0.7127612 0.6075414
## Lag 4096 0.3426658 0.6579238 0.5308334
## Lag 5120 0.2897890 0.6093251 0.4626872
## nodefactor.agecat.3 nodefactor.agecat.4 absdiff.sqrtage
## Lag 0 1.0000000 1.0000000 1.0000000
## Lag 1024 0.7238485 0.5948038 0.4254256
## Lag 2048 0.5552313 0.3990582 0.2356761
## Lag 3072 0.4437020 0.2789888 0.1697984
## Lag 4096 0.3640360 0.1960868 0.1373669
## Lag 5120 0.2958634 0.1422103 0.0945788
## concurrent
## Lag 0 1.0000000
## Lag 1024 0.4966418
## Lag 2048 0.3078943
## Lag 3072 0.2249880
## Lag 4096 0.1757426
## Lag 5120 0.1614681
##
## Sample statistics burn-in diagnostic (Geweke):
## Chain 1
##
## Fraction in 1st window = 0.1
## Fraction in 2nd window = 0.5
##
## edges nodefactor.agecat.1 nodefactor.agecat.2
## -1.4246 -2.9090 0.1335
## nodefactor.agecat.3 nodefactor.agecat.4 absdiff.sqrtage
## 0.9711 -1.4368 -1.6917
## concurrent
## -1.4674
##
## Individual P-values (lower = worse):
## edges nodefactor.agecat.1 nodefactor.agecat.2
## 0.154270565 0.003626204 0.893764616
## nodefactor.agecat.3 nodefactor.agecat.4 absdiff.sqrtage
## 0.331477249 0.150788470 0.090708972
## concurrent
## 0.142268348
## Joint P-value (lower = worse): 0.06146616 .
##
## MCMC diagnostics shown here are from the last round of simulation, prior to computation of final parameter estimates. Because the final estimates are refinements of those used for this simulation run, these diagnostics may understate model performance. To directly assess the performance of the final model on in-model statistics, please use the GOF command: gof(ergmFitObject, GOF=~model).
NetDX
MARCOH 10,000 node netwoks
Nodecov(age)
Static
##
## Network Diagnostics
## -----------------------
## - Simulating 10000 networks
## - Calculating formation statistics## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Static
## Simulations: 10000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 1140.799 1139.403 -0.122 20.787
## nodecov.age 74493.142 74363.184 -0.174 1334.808
## absdiff.sqrtage 305.576 307.618 0.668 10.641
## offset(nodematch.male) NA 0.000 NA 0.000
## offset(concurrent) NA 0.000 NA 0.000
## meandeg NA 0.455 NA 0.008
## concurrent NA 0.000 NA 0.000
Dynamic
##
## Network Diagnostics
## -----------------------
## - Simulating 5 networks
## |*****|
## - Calculating formation statistics
## - Calculating duration statistics
## |*****|
## - Calculating dissolution statistics
## |*****|
## ## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Dynamic
## Simulations: 5
## Time Steps per Sim: 5000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 1140.799 1126.836 -1.224 18.353
## meandeg NA 0.450 NA 0.007
## concurrent NA 0.000 NA 0.000
## nodecov.age 74493.142 73589.759 -1.213 1210.918
## absdiff.sqrtage 305.576 303.219 -0.771 10.331
##
## Dissolution Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## Edge Duration 463.528 418.056 -9.810 413.278
## Pct Edges Diss 0.002 0.002 -0.351 0.001
Nodefactor(agecat)
Static
##
## Network Diagnostics
## -----------------------
## - Simulating 10000 networks
## - Calculating formation statistics## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Static
## Simulations: 10000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 1140.799 1140.027 -0.068 17.557
## nodefactor.agecat.1 31.615 31.875 0.822 6.284
## nodefactor.agecat.2 259.360 258.316 -0.403 14.998
## nodefactor.agecat.3 486.403 484.052 -0.483 16.221
## absdiff.sqrtage 305.576 306.268 0.227 10.055
## offset(nodematch.male) NA 0.000 NA 0.000
## offset(concurrent) NA 0.000 NA 0.000
## meandeg NA 0.455 NA 0.007
## concurrent NA 0.000 NA 0.000
Dynamic
##
## Network Diagnostics
## -----------------------
## - Simulating 5 networks
## |*****|
## - Calculating formation statistics
## - Calculating duration statistics
## |*****|
## - Calculating dissolution statistics
## |*****|
## ## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Dynamic
## Simulations: 5
## Time Steps per Sim: 5000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 1140.799 1127.615 -1.156 19.219
## meandeg NA 0.451 NA 0.008
## concurrent NA 0.000 NA 0.000
## nodefactor.agecat.1 31.615 31.906 0.920 7.233
## nodefactor.agecat.2 259.360 254.924 -1.710 13.671
## nodefactor.agecat.3 486.403 479.107 -1.500 15.363
## absdiff.sqrtage 305.576 299.681 -1.929 10.692
##
## Dissolution Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## Edge Duration 463.528 414.706 -10.533 407.002
## Pct Edges Diss 0.002 0.002 0.196 0.001
OTHER 10,000 node networks
Nodecov(age)
Static
##
## Network Diagnostics
## -----------------------
## - Simulating 10000 networks
## - Calculating formation statistics## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Static
## Simulations: 10000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 398.805 398.420 -0.097 17.941
## nodecov.age 20625.627 20599.740 -0.126 986.277
## absdiff.sqrtage 115.858 115.919 0.052 7.753
## concurrent 32.831 32.794 -0.112 5.997
## offset(nodematch.male) NA 0.000 NA 0.000
## meandeg NA 0.159 NA 0.007
Dynamic
##
## Network Diagnostics
## -----------------------
## - Simulating 5 networks
## |*****|
## - Calculating formation statistics
## - Calculating duration statistics
## |*****|
## - Calculating dissolution statistics
## |*****|
## ## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Dynamic
## Simulations: 5
## Time Steps per Sim: 5000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 398.805 398.841 0.009 17.396
## meandeg NA 0.159 NA 0.007
## concurrent 32.831 33.098 0.816 6.237
## nodecov.age 20625.627 20630.129 0.022 967.800
## absdiff.sqrtage 115.858 115.627 -0.200 7.420
##
## Dissolution Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## Edge Duration 94.848 93.137 -1.804 92.597
## Pct Edges Diss 0.011 0.011 -0.175 0.005
Nodefactor(agecat)
Static
##
## Network Diagnostics
## -----------------------
## - Simulating 10000 networks
## - Calculating formation statistics## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Static
## Simulations: 10000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 398.805 399.576 0.193 17.590
## nodefactor.agecat.1 167.087 165.801 -0.770 13.886
## nodefactor.agecat.2 251.906 252.045 0.055 17.391
## nodefactor.agecat.3 156.357 158.498 1.369 13.852
## nodefactor.agecat.4 98.790 99.182 0.397 11.333
## absdiff.sqrtage 115.858 116.566 0.611 7.835
## concurrent 32.831 33.196 1.113 6.161
## offset(nodematch.male) NA 0.000 NA 0.000
## meandeg NA 0.160 NA 0.007
#### Dynamic
##
## Network Diagnostics
## -----------------------
## - Simulating 5 networks
## |*****|
## - Calculating formation statistics
## - Calculating duration statistics
## |*****|
## - Calculating dissolution statistics
## |*****|
## ## EpiModel Network Diagnostics
## =======================
## Diagnostic Method: Dynamic
## Simulations: 5
## Time Steps per Sim: 5000
##
## Formation Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## edges 398.805 401.428 0.658 18.401
## meandeg NA 0.160 NA 0.007
## concurrent 32.831 33.620 2.404 6.135
## nodefactor.agecat.1 167.087 165.607 -0.886 13.730
## nodefactor.agecat.2 251.906 253.866 0.778 17.796
## nodefactor.agecat.3 156.357 158.154 1.149 13.461
## absdiff.sqrtage 115.858 117.287 1.233 8.382
## nodefactor.agecat.4 98.790 NA NA NA
##
## Dissolution Diagnostics
## -----------------------
## Target Sim Mean Pct Diff Sim SD
## Edge Duration 94.848 93.178 -1.760 92.464
## Pct Edges Diss 0.011 0.011 -0.201 0.005
